NEAR-INFRARED SPECTROSCOPY IN FOOD ANALYSIS

1
in the spectra caused by physical properties such as the particle size of powders. This means that NIR spectroscopy becomes a secondary method requiring calibration against a reference method for the constituent of interest. As a consequence of the physics of diffuse transmittance and reectance and the complexity of the spectra, calibration is normally carried out using multivariate mathematics (chemometrics). NIR spectroscopy is used routinely for the compositional, functional and sensory analysis of food ingredients, process intermediates and nal products. The major advantage of NIR is that usually no sample preparation is necessary, hence the analysis is very simple and very fast (between 15 and 90 s) and can be carried out on-line. One of the strengths of NIR technology is that it allows several constituents to be measured concurrently. In addition, for each fundamental vibration there exists a corresponding series of overtone and combination bands with each successive overtone band approximately an order of magnitude less intense than the preceding one. This provides a built-in dilution series which allows several choices of absorptions of different intensity containing the same chemical information. Finally, the relatively weak absorption due to water enables high-moisture foods to be analyzed. The major limitation of NIR spectroscopy in food analysis is its dependence on less-precise reference methods.

Near-infrared Spectroscopy in Food Analysis

Brian G. Osborne BRI Australia Ltd, North Ryde, Australia

1 Introduction 2 Instrumentation 2.1 Monochromators 2.2 Diode Array Spectrometers 2.3 Filter Instruments 3 Sample Presentation 3.1 Diffuse Transmittance 3.2 Diffuse Reectance 3.3 On-line Samplers 4 Calibration Development 4.1 Data Preprocessing 4.2 Sample Population Structuring 4.3 Regression Techniques 4.4 Qualitative Analysis 5 Food Applications 5.1 Cereals and Cereal Products 5.2 Milk and Dairy Products 5.3 Meat 5.4 Fish 5.5 Fruit and Vegetables 5.6 Confectionery 5.7 Beverages 5.8 Authenticity 6 The Literature 7 Standard Methods Acknowledgments Abbreviations and Acronyms Related Articles References

1 2 3 3 3 3 3 4 4 6 6 6 6 7 7 7 10 10 10 10 11 11 11 11 12 12 12 13 13

1 INTRODUCTION
The NIR forms that part of the electromagnetic spectrum in the wavelength range 780 nm to 2500 nm (Figure 1). NIR, like all radiation, behaves as a wave with the properties of simple harmonic motion which may be dened in terms of two properties: the frequency of vibration D no. of times the wave pattern is repeated in 1 s D angular velocity/2 the wavelength D velocity of light frequency

Near-infrared (NIR) spectroscopy is based on the absorption of electromagnetic radiation at wavelengths in the range 780 2500 nm. NIR spectra of foods comprise broad bands arising from overlapping absorptions corresponding mainly to overtones and combinations of vibrational modes involving C H, O H and N H chemical bonds. The concentrations of constituents such as water, protein, fat and carbohydrate can in principle be determined using classical absorption spectroscopy. However, for most food samples, this chemical information is obscured by changes

Chemical bonds between atoms in molecules vibrate and to a rst approximation this vibration behaves as a simple harmonic motion. The motion of each atom may be treated as an independent vibration with respect to a xed centre of mass of the molecule rather like a mass attached to a spring. The vibration frequency is a function of the masses of the two atoms m1 and m2 and the strength of the bond k and there is a parabolic relationship between potential energy and interatomic distance.

Encyclopedia of Analytical Chemistry Edited by Robert A. Meyers. John Wiley & Sons Ltd, Chichester. ISBN 0471 97670 9

2
Wavelength (nm) 15 000 IR 2500 1800 NIR 750 Visible 400 1300 UV 190 800 1100 CH 2nd overtone OH combination NH 2nd overtone OH 2nd overtone CH 3rd overtone 1420 1600 CH 1st overtone OH, NH 1st overtone CH combinations 2500 2200 CH combinations OH, NH combinations

FOOD

Figure 1 Principal types of NIR absorption bands and their

locations.
0.8

0.6

0.4

0.2

0.0 1100 1300

1500 1700 1900 2100 Wavelength (nm)

2300

2500

Figure 2 A typical spectrum of a powdered cereal sample

(upper curve) and some of its constituent absorptions.

When the frequency of the radiation matches that of the vibrating molecule, there will be a net transfer of energy from the radiation to the molecule which can be measured as a plot of energy versus wavelength called a spectrum (Figure 2). However, energy is not transferred continuously but in discrete packets called quanta. In consequence, molecules can only occupy discrete energy levels dened by whole numbers 0, 1, 2, . . . and the energy change required for a transition between two levels is directly related to the frequency of the energy by multiplying by h (Plancks constant). For simple harmonic motion, the absorption or emission of energy can only take place if it matches a transition between any pair of adjacent energy levels. Since almost all the molecules in a sample will normally occupy the lowest energy level 0, this means that the only possible transition is from 0 to 1, called the fundamental. The simple harmonic model does not adequately describe the vibrational motion of chemical bonds

because at one extreme atoms repel one another the closer they approach and at the other extreme the molecule dissociates. For this more realistic model, transitions from energy level 0 to 2, 0 to 3 etc. become possible and these are referred to as rst, second etc. overtones with frequencies corresponding approximately to twice, three times etc. that of the fundamental. The frequencies of many overtone bands are in the NIR region. The greater the deviation of a particular bond system from harmonicity, the greater the intensity of overtone bands and the most anharmonic bonds are those involving the lightest atom, hydrogen, and a heavier atom such as carbon, nitrogen or oxygen. While each frequency of vibration of a complex molecule generally involves oscillation of all the atoms, those atoms undergoing large amplitude motions are often limited to a particular functional group, e.g. OH, NH2 etc. The characteristic frequency is similar for the same functional group where it appears in different molecules. This concept of characteristic group frequencies is the basis of analysis by infrared (IR) and, because each fundamental will be associated with its own potential energy curve and its own set of overtones, of analysis by NIR also. The number of bands a given molecule will produce can be calculated from the number of coordinates required to describe the total momentum of the molecule. 3n coordinates describe the location in space of a molecule of n atoms of which 3 describe the translational and 3 the vibrational motion (2 if linear) of the molecule. There are therefore 3n 6 (3n 5 if linear) possible modes of vibration but some of these may be the same. Furthermore, absorption can only occur if the molecular vibration is accompanied by a change in dipole moment such as with the hydrogen chloride molecule. Another type of absorption band occurs in the NIR region. Combination bands arise by interaction of two or more vibrations taking place simultaneously and the frequency of a combination band is simply the sum of multiples of the relevant fundamental frequencies. Thus, a triatomic molecule with 3 fundamental bands could have 3 rst overtones, 3 second overtones etc. and 6 combination bands. It soon becomes obvious that NIR spectra are much more complex than IR spectra. Nevertheless, overtones and combinations can still be assigned to group frequencies.

Log (1/R)

2 INSTRUMENTATION
Successful application of NIR depends on the correct choice of instrument. There are three different types available (described in detail by Osborne et al. 1 )

NEAR-INFRARED SPECTROSCOPY IN FOOD ANALYSIS

3
techniques. There is a technique available for any type of liquid, slurry, powdered or solid sample. 3.1 Diffuse Transmittance Radiation interacting with a sample may be absorbed, transmitted or reected. In the classical spectroscopy experiment, reection is eliminated so that the proportion of radiation attenuated by the sample may be measured as transmittance. Beers law then denes a proportionality between transmittance and the product of concentration of the absorbing species and path length. For a clear transparent liquid sample such as beer, hot starch melts, wine or vegetable oil, the path length may be xed by means of a static or ow-through sample cuvette or a pair of ber-optic probes and a calibration developed using samples of known concentrations. For example, Halsey 2 used standard solutions of ethanol in water to develop a calibration for alcohol content of beer. It should be noted that, owing to the relatively weak intensities of NIR absorption bands, samples such as vegetable oils may be analyzed without dilution in a solvent. Beers law is only valid in the absence of light scatter in the sample. Scattering changes the path length through which the radiation passes and, because the amount of scattering varies from sample to sample, the path length cannot be dened. This type of experiment is known as diffuse transmittance (Figure 3), the most wellknown example of which is liquid whole milk. The fat globules in the milk scatter light in the manner shown and invalidate Beers law. Diffuse transmittance measurements are usually carried out in the 800 1100 nm region of the spectrum where the weak absorptions enable useful data to be obtained using thicknesses of 1 2 cm of samples such as meat, cheese or whole grain. Nearinfrared transmittance (NIT) instruments are particularly applicable to the analysis of whole grains and a typical apparatus is shown schematically in Figure 4. A sample of grain is placed in a hopper from where aliquots are dispensed into the measurement chamber. When analysis is complete, the sample is discharged into a collection tray. This arrangement lends itself well to adaptation for on-line measurements (see section 5.1).

2.1 Monochromators Grating monochromators are used to measure the full visible and NIR spectrum and may be used in transmittance or reectance mode. They are therefore the most versatile instruments. Monochromators are generally used for research or when a wide range of different applications is required. Three different detectors are available: a silicon detector covers the range 400 1100 nm, an indium gallium arsenide covers the range 800 1700 nm and a lead sulde the range 1100 2500 nm. Some instruments contain both silicon and lead sulde detectors (giving them a wavelength range 400 2500 nm), some only the lead sulde. The latter would be restricted to measurement of powdered or granular samples by diffuse reectance (see section 3.2) or transmittance via ber-optic probes. At least one model of NIR monochromator has been designed for on-line use via either ber-optics or a powder sampler. Another type of dispersive monochromator used in NIR instruments is the acousto-optically tunable lter (AOTF). An AOTF comprises a crystal of TeO2 through which a plane travelling acoustic wave is generated at right angles to the incident light beam. This causes the crystal to behave as a longitudinal diffraction grating with a periodicity equal to the wavelength of sound across the material. The main advantages of AOTF over grating instruments is their mechanical simplicity (i.e. no moving parts) and their wavelength stability. 2.2 Diode Array Spectrometers Diode array spectrometers employ an array of IRemitting diodes. These function as both the light source and the wavelength selection system. Diode array instruments typically cover the range 400 1700 nm. They have the advantages that the measurement is very fast (e.g. one spectrum per second) and noninvasive. These features are particularly useful where a high sample throughput or ultra-rapid on-line measurements are required. 2.3 Filter Instruments The simplest and cheapest NIR instruments are based on a limited number, usually between six and twenty, of interference lters. These lters are chosen to represent the absorptions used for the most popular applications, e.g. protein, moisture and oil in agricultural samples. Filter instruments are designed for a limited range of routine analyses, either in the laboratory or on-line.

Detector Source

3 SAMPLE PRESENTATION
Lens Sample Cell

The wide range of applications of NIR in food analysis is possible because of the different sample presentation

Figure 3 Diffuse transmittance spectroscopy.

FOOD

Figure 4 Diffuse transmittance spectroscopy of a whole grain

sample.

3.2 Diffuse Reectance For a smooth surface such as glass, most of the radiation is reected from the surface by regular or specular reection and no absorption takes place. In the 1100 2500 nm region, the amount of scattering makes the path length so great that transmittance through 1 cm of most samples is negligible. This situation is called diffuse reectance because most of the incident radiation is reected. If a matt surface reects diffusely without penetration into the sample, like specular reectance no absorption takes place. If, however, some of the radiation penetrates the surface when it reaches each particle it can be reected, absorbed or transmitted. The net result is that the diffusely reected radiation (R) can be empirically related to concentration (c) in an analogous way to Beers law i.e. log 1/R D kc where k is a factor which incorporates both absorbtivity and path length. A typical NIR diffuse reectance experiment is shown in Figure 5. A powdered sample such as wheat meal is

packed into a 1 cm-deep sample cell and covered with a quartz window. The sample cell is then placed into the instrument where it is illuminated with NIR radiation and the reected radiation measured by either a set of detectors set at 45 to the incident beam or a single detector onto which the radiation is focused by means of an integrating sphere. This conguration has been adapted for on-line measurement using a powder sampler for our or milk powders. The diffuse reectance experiment can be adapted for use with liquids by placing a ceramic tile beneath the sample. In this sample presentation mode, known as transectance, the radiation is transmitted through the sample, reected from the ceramic then transmitted back through the sample before nally reaching the detector. Transectance is thus a hybrid of transmittance and reectance. Interactance, another hybrid of transmittance and reectance, involves illumination and detection at laterally separated points on the samples surface. It is normally accomplished using a ber-optic probe in which one set of ber-optic bundles carries the incident radiation and another carries the reected radiation. This type of arrangement is particularly useful for large samples such as intact fruit. 3.3 On-line Samplers Ideally, a sensor needs to be sufciently compact and robust to be placed in the required position on-line. It needs to be relatively low cost, rapid and capable of measuring the desired analyte or property on an intact sample. Finally, it must be capable of accurate and stable calibration. NIR instruments meet these requirements. Since the applications are limited and predened, lter instruments which full the criteria of compactness, robustness and cost can usually be employed. NIR is inherently a rapid technique (as fast as one spectrum per second with diode array technology) and data from many measurements can be combined into a moving average so as to identify trends. There are three types of NIR on-line analyzer. 3.3.1 Remote (Noncontact) Sensor

NIR Beam R R

Glass window

Sample

Figure 5 Diffuse reectance analysis of a powdered sample.

The rst dedicated on-line NIR sensor was invented by Edgar and Hindle of Infrared Engineering Co in the UK. This is based on a sensing head, which is referred to as a gauge, remote from the sample. This design has the advantages of a low cost of instrument and simplicity of installation but imposes severe constraints on instrument design as it is susceptible to interference from ambient light variations, dust build-up on the optical surfaces and atmospheric humidity variation. The Infrared Engineering gauge was therefore designed specically for on-line application and constructed to

NEAR-INFRARED SPECTROSCOPY IN FOOD ANALYSIS

5
presentation was selected on the principle that the sample needed to be held stationary against the optics window during the measurement cycle and that the our surface should be smooth, continuous and rmed under pressure. These conditions were met by taking direct samples from the our stream, either from gravity-fed spouting or a positive pressure blowline, into a mechanism designed to reproduce the action of a human operator in packing our into a sample cup. NIR on-line samplers were rst developed to measure the protein content of our. This is still the most popular application and provides an excellent example of an NIR feedback control system. Dried gluten is commonly used, particularly in Europe, to replace wholly or partly the protein in our which would otherwise be derived from high-protein wheat in the grist. The success of on-line NIR for monitoring our protein content therefore led to its incorporation into a closed-loop control system for gluten addition to our. 4 A mixing screw is installed between the gluten feed and the NIR sampler station from which a feedback signal controls the gluten feeder. The system has proved to be an efcient and accurate method of control of gluten addition to achieve a target protein content in the our. 3.3.3 Fiber-optic Probe Fiber-optic probes have the widest range of applications in on-line food analysis. Uses include dairy products, meat, fruit, beer and extruded snack foods. Extrusion cooking is used to manufacture a wide range of products from cereals, including breakfast cereals, snack foods and pet foods. In these products, texture and density are key quality requirements which are controlled by the processing conditions summarized by the term degree of cook. Conventionally, control is attempted by measurement of inputs such as the moisture content, screw speed, barrel temperature, etc. on the basis that the relationship between these and the product quality is understood. However, a more direct means of control is highly desirable. In contrast to breadmaking, where the product structure is largely controlled by protein development, in extrusion cooking it is the starch component of the cereal-based raw material which is the chief functional ingredient. Therefore, the aim was to use NIR to monitor changes in starch structure. Initially, products were prepared over a wide range of processing conditions and the extrudates freeze-dried and ground to powders prior to NIR spectroscopy. Since successful calibrations based on spectral features related to starch structure were obtained using the powdered samples, 5 the measurement was attempted on-line. A ber-optic reectance probe was installed in the die of the extruder and the light was transmitted through

minimize such potential interferences. The Infrared Engineering MM55 Gauge is based on a non-contacting sensing head xed approximately 200 mm from the product ow by mounting to the side of a section of spouting inclined at an angle of 60 from the horizontal. The gauge focuses light from a quartz halogen lamp into a parallel beam which is projected onto the product via a lter wheel which transmits narrow bands of energy at the desired NIR wavelengths. The reected energy, which has been modulated by interaction with the product, is captured by a detector in the gauge. In this way, measurements on the moving our stream are taken through a toughened glass window at a rate of up to ve per second. Signals from the MM55 gauge are conveyed to an electronic processing unit which contains the calibration equation and produces the analytical result. This is passed to a strip recorder to provide a permanent record. Applications of the MM55 in the food industry include moisture determination on both continuous samples such as powders and discontinuous samples such as whole biscuits, control of protein in our, monitoring of fat content of potato crisps and sorting of fruit. Recently, in Australia, a novel application of NIR spectroscopy to the monitoring of bread dough development has been invented by Wesley et al. 3 The invention makes use of the high scan speed (one spectrum per second) of the Perten DA-7000 diode array instrument. In this case, the instrument was inverted over an open mixing bowl and the signal from the dough was recorded continuously without stopping the mixer. In this way, using second derivative data at dened wavelengths assigned to protein and water, an NIR mixing curve was obtained. The NIR mixing curves were in good agreement with curves obtained using measurement of mixer torque over a range of our types and mixer conditions. 3.3.2 By-pass Sampler To control the composition of the grist (wheat mixture) during the blending process, it is necessary to perform measurements on the grain itself because the time delay between blending the wheats and producing the nal our is too long for an effective feedback control loop. Whole grain NIR instruments generally employ a hopper and ow-through cell and are thus relatively easy to adapt to on-line use. An on-line version of the Infratec 1225 Whole Grain Analyzer is commercially available. One of the rst near-infrared on-line powder samplers (NIROS) was based on an InfraAlyzer 300B instrument which was split into two parts comprising the electronics unit which would be removed from the immediate vicinity of sampling and the optical unit which was to be modied for on-line sample presentation. This method of sample

6
the melt and reected back from a polished steel pillar; in effect, therefore, a transectance measurement was used. The spectral characteristics of the different forms of starch found in the hot melts close to the die of the extruder were identical to those of the powdered extrudates. This research has opened the way to an online process monitoring and control system for extrusion cooking.

FOOD

(2, 8, 8, 1) means a second derivative with a 16 nm gap and 16 nm segment. Derivatives have two useful properties: 1. 2. resolution of overlapping bands; deconvolution of background (derivative ratios 6 correct for multiplicative effects).

4 CALIBRATION DEVELOPMENT
In NIR analysis, the calibration provides all three basic functions of analytical chemistry i.e. separation, identication and quantication. Therefore, derivation of the calibration equation for each application is the most crucial step in reliable and accurate analysis. There is, however, no unique method for deriving satisfactory calibration equations. The following sections provide an overview of the main techniques used and more detail can be found in Osborne et al. 1 4.1 Data Preprocessing Quantitatively the most important factor in the NIR reectance spectrum is the particle size of the sample. Log 1/R increases with increasing particle size because the apparent path length becomes longer. Therefore, for example, the same wheat sample ground to different particle sizes will result in substantially different spectra. However, the effect is not additive but multiplicative, i.e. proportional to log 1/R . Since in reectance spectra of foods and agricultural materials log 1/R coincidentally increases with increasing wavelength, the effect of particle size appears to be a function of wavelength. 6,7 Correction for particle size is accomplished empirically by multiple regression mathematics but many spectroscopists prefer to use derivatives. The rst derivative d log 1/R dl is the slope of the spectrum at wavelength l and is calculated as the difference between log 1/R at two adjacent wavelengths or, more usually, two segments in which the data are smoothed (averaged) over a small region of the spectrum dened by the segment size. The difference between these wavelengths or segments is called the gap. The second derivative d2 log 1/R dl2 is the difference between two adjacent rst derivatives. Derivative mathematics is described using the notation (derivative order, gap, segment, second smoothing) e.g.

The effect of particle size can also be overcome by multiplicative scatter correction 8 or standard normal variate and detrend. 9 The multiplicative scatter correction algorithm proceeds by computing the mean spectrum of the sample set and rotating each sample spectrum so that a regression line drawn through it aligns with a regression line drawn through the mean spectrum. 4.2 Sample Population Structuring The NIR analysis of food and agricultural samples is based on calibration against a reference method using a set of samples which represent all the variability in the population from which they are drawn. The problem becomes one of identifying these samples. One set of criteria would be to obtain analytical data on a large database of samples and select those with the most variation in composition. However, this is an expensive option. The concept of population structuring was introduced by Shenk and Westerhaus. 10 The idea is to assemble a library containing the NIR spectra of a large number of samples and use discriminant analysis techniques to dene the boundaries of the library. The spectra of further samples can then be examined to ascertain whether they belong to the population represented by the library and whether their spectra are different to those of existing samples. The spectral library can be used in several ways: to dene the optimum calibration set; to verify that test samples match the calibration; to produce locally weighted calibrations for optimum measurement accuracy; to identify new samples with which to extend the calibration.

4.3 Regression Techniques The usual method of calibration of NIR instruments is by multiple linear regression of the reference data on the spectral data. The difculty in this is twofold. Firstly, NIR spectra contain a large amount of data whereas only a few terms should be included in a regression equation. Secondly, because of the effect of scatter, the raw NIR data are highly intercorrelated and this precludes the use of forward stepwise regression,

NEAR-INFRARED SPECTROSCOPY IN FOOD ANALYSIS

7
17

1.0

0.5

15

Correlation

0
13

0.5 1.0 2000

NIR
11

2080

2160 Wavelength (nm)

2240

2320
9

Figure 6 Plot of correlation versus wavelength for a set of

calibration samples.
7 7 9 11 Reference 13 15 17

the most common method of selection of terms in multiple linear regression. For xed lter instruments, the wavelengths for each application may be known. If not, there are only a few from which to select and is feasible to employ an all possible combinations strategy. For monochromator or diode array instruments on the other hand, more sophisticated methods are necessary. The use of mathematical pretreatments such as derivatives or multiplicative scatter correction (see section 4.1) would make forward stepwise selection more reliable but it is important to ensure that the selected wavelengths are consistent with known chemical absorptions. This can be done by comparing a plot of correlation versus wavelengths (Figure 6) with the spectrum of the constituent of interest. 11 However, this is not always possible. For example, functional properties do not usually have a dened spectral pattern. An increasingly popular approach, therefore, is to reduce the amount of data by principal components analysis then perform linear regression on the principal components. If the principal components are selected with regard to minimizing the correlation coefcient, the method becomes partial least squares (PLS). The end result is a calibration equation from which the constituent of interest may be calculated from a (usually) linear combination of spectral data. The equation has associated statistics which dene the closeness of t of the actual and predicted values to the least squares line. It is always important to plot a scatter graph as illustrated in Figure 7 in order to detect any aberrant data. Ideally, the scatter plot should contain data points distributed evenly about the line but within the condence limits as shown in the example. 4.4 Qualitative Analysis In applications such as authenticity testing (see section 5.8), where the aim is to classify rather than perform a

Figure 7 A typical scatter plot of NIR versus reference data. quantitative determination, a qualitative or discriminant analysis approach is often used. The mathematical approach to this problem is to develop a model which denes the mean and standard deviation of each sample type in multidimensional space then test to which group an unknown sample belongs. This test is based on a nonEuclidean distance measurement such as the Mahalanobis distance. The Mahalanobis distance is dened according to elliptical contours spreading out from the group centres. The mathematics and applications of discriminant analysis are described by Mark. 12 More sophisticated classication models have been achieved using articial neural networks (ANNs). ANNs consist of input nodes which receive one piece of spectral data each and distribute it to a hidden layer of nodes where the data are transformed before being distributed again to a set of output nodes. The nodes simply provide the mechanism for distribution of the data through the hidden layer which may be considered a nonlinear function approximation machine. The crucial feature of ANNs is that they are adaptive. One example of the application of ANNs to a food classication problem has been published by Hervas et al. 13

5 FOOD APPLICATIONS
5.1 Cereals and Cereal Products NIR has been used for the quality testing of crossbred material from wheat breeding programs since the late 1970s. Because NIR is rapid, low cost per test and requires a relatively small quantity of sample, it is routinely used

8
Table 1 NIR applications in quality testing of
wheat breeding lines Constituent Protein Moisture Hardness Water absorption Wheat Flour

FOOD

(indicative)

throughout the world to determine the protein content of wheat to allow breeders to screen large numbers of lines for this key characteristic. In addition, NIR is widely used to replace some of the chemical tests necessary as part of a quality testing program (Table 1). In particular, NIR hardness and moisture measurements may be used to determine the conditioning requirements prior to test milling while protein and moisture may be determined on the resulting our. The ultimate application of NIR analysis in a wheat breeding program would be direct prediction of functional quality including our yield, damaged starch, water absorption, dough development time, extensibility and loaf volume. Although some breeding programs are using NIR to eliminate lines according to some of these measurements, they have not gained widespread acceptance. The reason appears to lie in the sensitivity of calibration to changes in variety, growing location and season, which necessitates frequent re-calibration. The availability to breeding programs of whole grain instruments provides a renewed incentive to develop the use of NIR to take the place of the entire procedure for quality screening of early generation lines where the numbers and quantity of samples preclude the use of the traditional methods. A recent development in instrumentation is the diode array spectrometer as exemplied by the Perten Model DA-7000. This type of instrument is quick to load as it requires no sample cell, and spectral data are acquired very rapidly; typically, an analytical result can be obtained in 15 s. NIR is used in Australia to predict optimum fertilizer requirements of cereal crops by analysis of total nitrogen and carbohydrate in plant tissue samples. The tissue test for rapid determination of shoot nitrogen status in cereal crops was rst developed for rice. The test has since been extended to encompass wheat shoot nitrogen and fructans. The tissue testing system is based on a plant sample taken by the farmer at a specic growth stage (Zadok scale) which is dried and analyzed by NIR. The nitrogen and fructan results are then entered into a database which is used to determine the appropriate fertilizer recommendation. 14 The use of NIR technology to help farm managers understand the fertility of their crops and to segregate or blend grain on its protein content prior to delivery

is gaining in popularity. In many countries, the price of wheat is determined by its protein content, often with substantial price increments between grades. This has encouraged some farmers to blend wheat on farm according to its NIR protein to increase their deliverable tonnage of higher priced grain. Providing the farms instrument has been properly calibrated and is well maintained, this strategy should be nancially successful. Some farmers are also realizing that NIR has the potential to analyze a large number of samples in a short period of time which could enable them to assess their crop for optimum development or to map differences in fertility across elds. NIR has been used in Australia, Canada, Europe and the USA to monitor growers deliveries of wheat and barley for two decades, and some of the rst research papers, e.g. Williams, 15 described the calibration of instruments for this purpose. Canada was the rst grainproducing country to utilize NIR for all wheat protein testing in its segregation program. As well as increased efciency of the Canadian wheat segregation program, the adoption of NIR testing has resulted in total cost savings to date of CAN$2.5 m. Furthermore, the replacement of the Kjeldahl system, which involves concentrated sulphuric acid and heavy metal catalyst, by the chemicalfree NIR method has enabled Canada to lead the way in environmentally safe large-scale protein testing of grain. The Canadian Grain Commission now uses NIT exclusively for providing quality data in support of the marketing of about 50 million tonnes of grain annually. By 1980, NIR protein testing had become the ofcial method of the US Federal Grain Inspection Service (FGIS) and since then has been in continuous routine use in wheat segregation in the USA. In the USA, following the Grain Quality Improvement Act 1986, hardness measurements have been introduced as part of the wheat grading system. Wheat hardness has been measured in both meal and whole grain. Since there is no international standard reference method for wheat hardness, a denition based on NIR spectra of ground grain has been proposed by Norris et al. 16 Delwiche et al. in the USA 17 have carried out a systematic study to develop a wheat classication system for US market class based on NIR protein and hardness measurements on single kernels. This concept is being developed by the FGIS into a practical system for the objective, rapid and sensitive detection of mixtures of wheat classes. In Europe in the early 1980s, our millers encountered a need to stimulate the cultivation of higher tonnages of high protein wheat by payment of premiums for protein content. A rapid method of load-by-load testing which produced results acceptable to both farmers and millers became necessary in order to calculate the price on

NEAR-INFRARED SPECTROSCOPY IN FOOD ANALYSIS

9
routine practice in our mills worldwide. It is used for testing each delivery of wheat in order to make decisions about acceptance, price and binning; for determination of conditioning time from measurement of hardness; and for analyzing our to check that it complies with specications before shipment to the customer. However, the power of NIR goes far beyond mere replacement of wet chemical analysis for raw material and product testing. With the introduction of an NIR instrument into the mill, the laboratory was able for the rst time to make a direct contribution to controlling the process in real time. So useful has NIR proved to be for this purpose that in many mills it soon became moved from the laboratory into the mill itself where it could be used off-line by shift millers for process monitoring whenever the mill is running. The success of NIR for this purpose gave it the necessary credibility as a potential on-line sensor in our mills (see section 3.3). Many large plant bakeries employ NIR to monitor the quality of their our and other raw ingredients. Typically, our would be tested for protein and moisture (possibly, but less commonly, color and water absorption). In biscuit and confectionery bakeries, other ingredients such as sucrose, chocolate etc. may also be analyzed. Having veried the composition of ingredients against their quality specications, the next point of control of a bakery process is at the dough mixing stage. Detection of faults in the dough avoids energy wastage in baking an out-ofspecication product and, for biscuit doughs, may allow the dough to be re-cycled. NIR has been used for compositional analysis of intact biscuit and bread dough pieces and for monitoring the progress of bread dough mixing. NIR calibrations for compositional analysis of shortsweet biscuit doughs enabled the simultaneous detection of errors in fat of 5% relative to the total amount of fat in the recipe and gross errors in the water, sucrose and our percentages. 20 NIR has also been used to determine the protein and fat contents in bread dough. During baking, the moisture content of doughs is reduced from about 18% to 2% for biscuits and 60% to 40% for bread. Monitoring the changes in the moisture content during baking allows the oven performance to be optimized. However, this is not easy because there is a non-uniform distribution of water content through both the dough and the freshly-baked product. For biscuits, an Infrared Engineering MM55 Gauge (see section 3.3) is commonly used to monitor the moisture contents of biscuits on a conveyer as they leave the oven. This measurement depends on a reproducible relationship between the gauge reading which represents the surface moisture content in the freshly-baked biscuit and the bulk moisture content of the fully equilibrated product in packet and on strategies to compensate for the discontinuous nature of discrete products on a conveyer.

delivery to the mill. NIR was adopted for this purpose in a number of European countries including UK, Ireland, Germany and Sweden. Initially, instruments based on reectance measurements on ground grain were used for testing grain at harvest and these were shown to result in calibrations for protein which were stable over growing locations and seasons and transferable between instruments. 18 By the mid 1980s, NIT instruments had become available. These instruments typically employ a hopper, into which a sample of whole grain is placed, to dispense the sample into the measurement compartment where it is analyzed by transmittance spectroscopy (Figure 4). Usually, 4 20 subsamples are analyzed sequentially and the results averaged to obtain a more representative result for the sample. The ability to perform protein testing on whole grain eliminates the need for grinding, thus simplifying and speeding up the analysis. Protein calibrations for whole wheat and barley have been shown to perform with equal accuracy to those derived for the ground grain. 19 However, the criteria for accuracy at receival are more stringent than those for selection of breeding lines. This is because grower payments are determined according to the NIR protein content. In some countries, the payment scale is not continuous but there are cutoffs between different grades where there may be substantial price differentials. Consequently, although the accuracy of NIR calibrations are normally assessed over a wide range, in this case the scatter around a much smaller range becomes important. In grain trading it is highly desirable to eliminate between-instrument variation in analysis results so that the farmer receives the correct grading and thus fair payment irrespective of the receival silo delivered to. This can be achieved by networking so that small differences in the response of individual instruments are corrected for in the calibration. All instruments in the network are connected to a central computer by modems, allowing simultaneous and uniform calibration and update of all instruments in the network. In 1995, all the European networks were brought together into The European Grain Network with the aim of standardizing NIR grain testing throughout the continent. Paddy rice must be harvested at the optimum maturity to obtain high yield and quality. This is assessed by the moisture content of rice at receival. NIR reectance instruments are unsuitable for use in rice receival testing because of the high moisture content at which rice is harvested. Following the successful development of NIT calibration for whole rice grain, Australian rice growers have adopted this technology for moisture testing on paddy rice. The use of NIR technology to determine the protein and moisture contents of both wheat and our is now

10
Table 2 NIR applications in the analysis of cereal products
Product Bread Biscuits Cake mixes Breakfast cereals Pasta Snack foods Moisture Protein Fat Starch Sugars Fiber

FOOD

NIR is applicable to the analysis of moisture, protein, fat, starch, sugars and ber in intact cereal foods such as bread, biscuits, cake mixes, breakfast cereals, pasta and snack foods (Table 2). NIR has been applied to the measurement of the rate of staling of white bread crumb without the use of a reference method. 21 Rate constants were calculated directly from plots of second derivative NIR data versus time. This is a rare example in food research where NIR has been used as a primary method. 5.2 Milk and Dairy Products Milk analyzers based on mid IR spectroscopy have been available since the 1960s but it was not until much later that NIR began to be used in the dairy industry. Although the mid IR is satisfactory for the proximate analysis of liquid milk, dairy chemists are faced with one of the widest ranges of sample types in the food industry. NIR has a key role in the analysis and process control of dairy products. It offers exibility in the analysis of protein, moisture, fat and lactose contents in a wide range of dairy products including: liquid milk dried whole milk, skim milk and whey powders cream traditional and processed cheese.

on-line NIR ber-optic measurement in dairy processing are moisture control of cream cheese and processed cheese. Cream cheese is manufactured in a series of set tanks from which product is transferred into a separator. NIR monitoring of the product from the separator enables compensation for the variation in characteristics from different set tanks. NIR measurements of processed cheese have been used for process diagnostics in which a greater understanding of the effect of temperature on the nal moisture content has been gained. The extent of development of NIR applications in the dairy industry is such that it is possible to purchase an instrument which is factory-calibrated for proximate analysis of cheddar cheese. However, these calibrations are only valid for cheddar made to the traditional recipe and sampled directly from the vat; non-traditional and matured cheddars require different calibrations. 5.3 Meat NIR spectroscopy is widely used in the meat industry for proximate analysis. A special interactance ber-optic probe has been designed to spear carcasses and determine their fat content. This enables the carcasses to be sorted prior to butchering. Dedicated instruments are available to determine the protein, fat and moisture contents of ground meat and meat products and factory-set calibrations are available for cooked meat, cooked ham, liver sausage, and pepperoni. Meat samples are minced then blended in a food processor before being packed into an open sample cell. The amount of sample needs to be controlled either by mass or by depth. Isaksson et al. 22 have used the MM55 on-line gauge (see section 3.3) for proximate analysis of ground beef directly at a meat grinder outlet. NIR has also been used for sensory analysis of sausages and to discriminate between fresh and frozen-then-thawed beef. 5.4 Fish The analysis of farmed atlantic salmon and rainbow trout enables the optimization of their processing. NIR ber-optic interactance probes can be used to measure nondestructively the protein, moisture and oil contents of whole sh. 23 5.5 Fruit and Vegetables Fresh fruit and vegetables are graded by shape, size and color. Objective, nondestructive methods of sorting enable growers and packers to market a consistent product over an extended season. In addition, high-quality produce can be segregated for high-return markets while

Many of these products are emulsions whose sampling for classical chemical analysis is difcult. For example, blending such samples changes their physical characteristics. NIR offers the possibility of on-line analysis which avoids the need for batch sampling and minimizes sampling error by averaging of virtually instantaneous, continuous measurements. Milk powders are analyzed on-line using a powder analyzer which enables the moisture content to be controlled. For most other on-line applications, ber-optic probes are used. These have the advantages of minimal maintenance owing to the absence of moving parts, robustness to high temperatures and pressures and the survival of clean-in-place protocols involving the use of caustic chemicals. Two examples of

NEAR-INFRARED SPECTROSCOPY IN FOOD ANALYSIS

11
Table 3 NIR applications in quality testing of barley breeding lines
Constituent Total nitrogen Moisture Lysine b-Glucan Extract Soluble protein Free a-amino N Barley Malt Wort

lower-quality produce can be identied for processing or other uses. The nondestructive sorting of fruit for ripeness by optical spectroscopy was originally carried out in the visible region. However, the availability of ber-optic interactance probes led to a resurgence of research interest in this application via the determination of sugar content. The reason for developing a direct measure of sugar content of fruit is that appearance is not a reliable guide to sweetness. Japanese researchers 24,25 have demonstrated the successful application of NIR to the on-line determination of sugar content in intact peaches and mandarins and developed an automated fruit sorting machine based on this principle. Australian scientists have extended this concept to tropical fruits such as melons, mangoes and pineapples. 5.6 Confectionery NIR applications in the confectionery industry include the determination of moisture in granulated sugar and chocolate crumb, protein and oil in cocoa powder and fat in whole chocolate. NIR data have also been shown to correlate highly with sensory data on raw and roasted cocoa beans, chocolate mass and nished chocolate. 5.7 Beverages When barley is germinated under controlled conditions, it undergoes a series of complex biochemical reactions which result in its conversion to malt. The malt is then mashed with water to produce a liquid called wort which is then fermented to produce beer. Since malting takes about twelve days, barley breeders need a rapid means of predicting the malting quality of barley grain. Calibrations have been developed for nitrogen, lysine, b-glucan and malt hot water extract. As with wheat functional quality, NIR analysis of barley is used to predict the quality of material which is derived by processing of the grain and there is doubt whether tests on ungerminated grain fully account for the interaction of enzymes and substrates during germination. Attempts have been made to improve the accuracy of prediction by analysis of key quality characteristics of the malt or the wort (Table 3). In the brewing industry, NIR is widely used to monitor the original gravity and alcohol content of beer using online ow-through cells. Standard errors of the calibration for alcohol of 0.1 0.2% have been reported 2 using either transmittance or transectance cells. The Liquidata probe has been used to achieve a process guarantee on any beer to better than 0.04% alcohol. NIR is used to monitor fruit quality and determine the alcohol content of wine and dedicated lter instruments for wine analysis are commercially available.

Other beverage applications include the moisture content of tea and coffee, sugar content of fruit juices and the sensory quality of tea and coffee blends. 5.8 Authenticity The study of food authenticity involves establishing whether a sample is genuine in terms of its description, including geographical origin. Many analytical techniques have been used to examine the authenticity of a wide range of different foods. The applications of NIR for authenticity testing of coffee, fruit pulps, milk powders, orange juice, pig carcasses, rice, sausages, sugars, vegetable oils, wheat grain and wheatour have been reviewed by Downey. 26 These applications are based on the principles of discriminant analysis (see section 4.4) in which the problem is to compare the spectrum of the test sample with a reference library (see section 4.2). The library should contain examples of known authentic and adulterated samples. In the simplest case, the question is simply whether the test sample belongs to one of two populations such as Basmati rice or not Basmati rice. Often, however, it is required to classify within a series of possible classes (e.g. different vegetable oils or wheat varieties), to identify a particular kind of adulteration (e.g. medium invert beet sugar, pulp wash or zest oil in orange juice) or to quantify an adulteration (e.g. non-dairy fat in milk products).

6 THE LITERATURE
Scientic papers describing the practical applications of NIR in food analysis are spread widely throughout the literature. As evidenced by the key references cited in this article, such papers may be found in journals devoted to analysis, spectroscopy, general food science or more specialized areas of food science. This makes it particularly difcult to search the literature. However, NIR News (NIR Publications, Chichester, UK) has since 1990 included a references section in an attempt to overcome this problem.

12
The only scientic journal devoted exclusively to NIR spectroscopy is Journal of Near Infrared Spectroscopy (NIR Publications, Chichester, UK). Volume 4 was dedicated to Karl H Norris, the founder of modern NIR analysis. It contains historical, review or leading edge papers by many of the pioneers of NIR spectroscopy and provides a particularly valuable introduction to the technique. The proceedings of the series of international conferences on near-infrared spectroscopy (ICNIRS) provide a valuable addition to the literature on the developments in and applications of NIR spectroscopy: J. Hollo, K. J. Kaffka, J. L. Gonczy, Near-Infrared Diffuse Reectance/Transmittance Spectroscopy, Akademiai Kiado, Budapest, 1987. C. S. Creaser, A. M. C. Davies, Analytical Applications of Spectroscopy, Royal Society of Chemistry, Cambridge, 1988. M. Iwamato, S. Kawano, Proceedings of the Second International Near-Infrared Spectroscopy Conference, Korin, Tokyo, 1989. R. Biston, N. Bartiaux-Thill, Proceedings of the Third International Conference on Near-Infrared Spectroscopy, Agricultural Research Centre Publishing, Gembloux, Belgium, 1990. K. I. Hildrum, T. Isakksson, T. Naes, A. Tandberg, Near-Infrared Spectroscopy. Bridging the Gap between Data Analysis and NIR Applications, Ellis Horwood, Chichester, 1992. G. D. Batten, P. C. Flinn, L. A. Welsh, A. B. Blakeney, Leaping Ahead with Near-Infrared Spectroscopy, Royal Australian Chemical Institute, Melbourne, 1995. A. M. C. Davies, P. C. Williams, Near-Infrared Spectroscopy: The Future Waves, NIR Publications, Chichester, 1995.

FOOD

Method 39-10. Near-infrared reectance method for protein determination. First Approved: 1982. Method 39-11. Near-infrared reectance method for protein wheat our. First Approved: 1985. Method 39-20. Near-infrared reectance method for protein and oil determination soybeans. First approved: 1991. Method 39-21. Near-infrared method for whole-grain analysis. First Approved: 1991. Method 39-25. Near-infrared method for protein content in whole-grain wheat. First Approved: 1995. Method 39-70A. Wheat hardness as determined by nearinfrared reectance. First Approved 1986. Association of Ofcial Analytical Chemists (AOAC) Ofcial First Action Method 997.06. Protein (crude) in wheat. Whole grain analysis, Near-infrared spectroscopic method. Adopted: 1997. International Association for Cereal Science & Technology (ICC) ICC Standard No. 159. Determination of protein by nearinfrared reectance (NIR) spectroscopy. Approved: 1995. ICC Recommendation No. 202. Procedure for nearinfrared (NIR) reectance analysis of ground wheat and milled wheat products. Approved: 1986. Royal Australian Chemical Institute (RACI) Cereal Chemistry Division Method No. 11.01 Determination of protein and moisture in whole wheat and barley by NIR. First Approved: 1998.

ACKNOWLEDGMENTS
The author thanks Mrs S. Ormston for preparation of the original artwork used in this article.

There are only two books currently in print devoted to NIR analysis: D. A. Burns, E. W. Ciurczak, Handbook of NearInfrared Analysis, Marcel Dekker Inc., New York, 1992. B. G. Osborne, T. Fearn, P. H. Hindle, Practical NIR Spectroscopy, 2nd edition, Longman, Harlow, 1993.

ABBREVIATIONS AND ACRONYMS

AOTF AACC ANN AOAC FGIS IR ICC ICNIRS NIR Acousto-optically Tunable Filter American Association of Cereal Chemists Articial Neural Network Association of Ofcial Analytical Chemists Federal Grain Inspection Service Infrared International Association for Cereal Science & Technology International Conferences on Near-Infrared Spectroscopy Near-Infrared

7 STANDARD METHODS
The following standard methods of NIR analysis of foods have been adopted: American Association of Cereal Chemists (AACC) Method 39-01. General near-infrared instrumentation and techniques. First Approved: 1989.

NEAR-INFRARED SPECTROSCOPY IN FOOD ANALYSIS

13
Spectra of Wheat, J. Near Infrared Spectrosc., 2, 93 99 (1994). P. Geladi, D. MacDougal, H. Martens, Linearization and Scatter Correction for Near-infrared Reectance Spectra of Meat, Appl. Spectrosc., 39, 491 500 (1985). R.J. Barnes, M.S. Dhanoa, S.J. Lister, Standard Normal Variate Transformation and De-trending of Nearinfrared Diffuse Reectance Spectra, Appl. Spectrosc., 43, 772 777 (1989). J.S. Shenk, M.O. Westerhaus, Population Structuring of Near Infrared Spectra and Modied Partial Least Squares Regression, Crop Sci., 31, 1548 1555 (1991). B.G. Osborne, Comparative Study of Methods of Linearization and Scatter Correction in Near Infrared Reectance Spectroscopy, Analyst, 113, 263 267 (1988). H. Mark, Qualitative Discriminant Analysis, in Handbook of Near-Infrared Analysis, eds. D.A. Burns, E.W. Ciurczak, Marcel Dekker, New York, 329 363, 1992. C. Hervas, A. Garrido, B. Lucena, N. Garcia, E. De Pedro, Near-Infrared Spectroscopy for Classication of Iberian Pig Carcasses Using an Articial Neural Network, J. Near Infrared Spectrosc., 2, 177 184 (1994). V.B. McGrath, A.B. Blakeney, G.D. Batten, Fructan to Nitrogen Ratio as an Indicator of Nutrient Status in Wheat Crops, New Phytol., 136, 145 152 (1997). P.C. Williams, Application of Near Infrared Reectance Spectroscopy to Analysis of Cereal Grains and Oilseeds, Cereal Chem., 52, 561 576 (1975). K.H. Norris, W.R. Hruschka, M.M. Bean, D.C. Slaughter, A Denition of Wheat Hardness Using NearInfrared Reectance Spectroscopy, Cereal Foods World, 34, 696 705 (1989). H. Song, S.R. Delwiche, Y.-R. Chen, Neural Network Classication of Wheat Using Single Kernel Near Infrared Transmittance Spectra, Opt. Engng., 34, 2927 2934 (1995). B.G. Osborne, T. Fearn, Collaborative Evaluation of Universal Calibrations for the Measurement of Protein and Moisture in Flour by Near Infrared Reectance, J. Food Technol., 18, 453 460 (1983). P.C. Williams, D.C. Sobering, Comparison of Commercial Near Infrared Transmittance and Reectance Instruments for Analysis of Whole Grains and Seeds, J. Near Infrared Spectrosc., 1, 25 32 (1993). B.G. Osborne, T. Fearn, A.R. Miller, S. Douglas, Application of Near Infrared Reectance Spectroscopy to the Compositional Analysis of Biscuits and Biscuit Dough, J. Sci. Food Agric., 35, 99 105 (1984). R.H. Wilson, B.J. Goodfellow, P.S. Belton, B.G. Osborne, G. Oliver, P.L. Russell, Comparison of Fourier Transform Mid Infrared Spectroscopy and Near Infrared Reectance Spectroscopy with Differential Scanning Calorimetry for the Study of the Staling of Bread, J. Sci. Food Agric., 54, 471 483 (1991).

NIROS NIT PLS RACI

Near-Infrared On-line Powder Sampler Near-Infrared Transmittance Partial Least Squares Royal Australian Chemical Institute

8.

9.

RELATED ARTICLES
Biomedical Spectroscopy Near Infrared Spectroscopy, In Vivo Tissue Analysis by Petroleum and Fossil Fuels Analysis Near Infrared Spectroscopy in Analysis of Crudes and Transportation Fuels Near Infrared Spectroscopy in Analysis of Petroleum Fractions and Additives Polymers and Rubbers Near-infrared Spectroscopy of Polymers and Rubbers Process Instrumental Methods Near Infrared Spectroscopy in Process Analysis Chemometrics Multivariate Calibration of Analytical Data General Articles Quantitative Spectroscopic Calibration
14. 10.

11.

12.

13.

15.

16.

REFERENCES
1. 2. B.G. Osborne, T. Fearn, P.H. Hindle, Practical NIR Spectroscopy, Longman, Harlow, 49 78, 1993. S.A. Halsey, The Use of Transmission and Transectance Near-Infrared Spectroscopy for the Analysis of Beer, J. Inst. Brew., 91, 306 312 (1985). I.J. Wesley, N. Larsen, B.G. Osborne, J.H. Skerritt, NonInvasive Monitoring of Dough Mixing by Near-Infrared spectroscopy, J. Cereal Sci., 27, 61 69 (1998). T. Fearn, P.I. Maris, An Application of Box Jenkins methodology to the Control of Gluten Addition in a Flour Mill, Appl. Statist., 40, 477 484 (1991). R.C.E. Guy, B.G. Osborne, P. Robert, The Application of Near-Infrared Reectance Spectroscopy to Measure the Degree of Processing in Extrusion Cooking Processes, J. Food Eng., 27, 241 258 (1996). K.H. Norris, P.C. Williams, Optimisation of Mathematical Treatments of Raw Near-infrared Signal in the Measurement of Protein in Hard Red Spring Wheat. I. Inuence of Particle Size, Cereal Chem., 61, 158 165 (1984). M. Manley, A.E.J. McGill, B.G. Osborne, The Effect of Light Scattering on NIR Reectance and Transmittance 17.

18.

3.

4.

19.

5.

20.

6.

21.

7.

14
22. T. Isaksson, B.N. Nilsen, G. Togersen, R.P. Hammond, K.I. Hildrum, On-line, Proximate Analysis of Ground Beef Directly at a Meat Grinder Outlet, Meat Sci., 43, 245 253 (1996). J.P. Wold, T. Isaksson, Non-destructive Determination of Fat and Moisture in Whole Atlantic Salmon by Near-infrared Diffuse Reectance Spectroscopy, J. Food Sci., 62, 734 736 (1997). S. Kawano, H. Watanabe, M. Iwamoto, Determination of Sugar Content in Intact Peaches by Near Infrared

FOOD

25.

23.

26.

24.

Spectroscopy with Fiber-Optics in Interactance Probe, J. Japan Soc. Hort. Sci., 61, 445 451 (1992). S. Kawano, T. Fijiwara, M. Iwamoto, Nondestructive Determination of Sugar Content in Satsuma Mandarin using Near Infrared (NIT) Transmittance, J. Japan Soc. Hort. Sci., 62, 465 470 (1993). G. Downey, Authentication of Food and Food Ingredients by Near Infrared Spectroscopy, J. Near Infrared Spectrosc., 4, 47 61 (1996).